This invention relates to a predictive coding device for use in carrying out prediction coding of a sequence of image data signals.
In general, a conventional predictive coding device of the type described is supplied with an image data signal sequence divided into a plurality of blocks and processes the image data signal sequence at every block by the use of efficient coding to produce a sequence of coded output signals. In this event, hybrid coding is very often used as such efficient coding so as to code the image data signal sequence into the coded output signals. In such hybrid coding, predictive coding is at first carried out to obtain a prediction error signal which is thereafter subjected to orthogonal transform. In addition, motion-compensated prediction is also used to obtain a motion vector in the conventional predictive coding device from time to time.
At any rate, processing is mainly finished or completed at every block in the conventional predictive coding device to obtain a result of processing, regardless of processing the other blocks adjacent to every block. For example, the orthogonal transform is carried out at every block and the motion vector is produced at every block.
Under the circumstances, it often happens that the results of processing are different from one another in adjacent ones of the blocks. Such differences bring about a block distortion which is visibly perceptible. Accordingly, a discontinuity is prone to occur at boundaries between the adjacent blocks and results in degradation of an image quality, when such a prediction error signal is reproduced.
In order to improve degradation of the image quality, consideration has been directed to an overlapping technique of blocks with one another so as to remove the discontinuity at the boundaries among the blocks. More specifically, a modulated discrete cosine transform (MDCT) and a lapped orthogonal transform (LOT) have been proposed as such an overlapping technique. In addition, subband coding has been also proposed which analyzes frequencies by the use of a filter.
However, it is difficult to solve the block distortion even when the transform is improved, as mentioned above. This is because block processing is carried out about predictive coding, motion-compensated prediction, and the like, irrespective of the other blocks.